📣 vSpeech 📜

V bindings for Mozilla's DeepSpeech TensorFlow based library for Speech-to-Text.

Install using vpkg

vpkg get https://github.com/thecodrr/vspeech

Install using V's builtin vpm (you will need to import the module with: import thecodrr.vspeech with this method of installation):

v install thecodrr.vspeech

Install using git:

cd path/to/your/project
git clone https://github.com/thecodrr/vspeech

You can use thecodrr.vave for reading WAV files.

Then in the wherever you want to use it:

import thecodrr.vspeech //OR simply vave depending on how you installed
// Optional
import thecodrr.vave

Perform the following steps:

1. Download the latest native_client.<your system>.tar.xz matching your system from DeepSpeech's Releases.

2. Extract the .tar.xz into your project directory in libs folder. It MUST be in the libs folder. If you don't have one, create it and extract into it.

3. Download pre-trained model from DeepSpeech's Releases (the file named deepspeech-0.6.1-models.tar.gz). It's pretty big (1.1G) so make sure you have the space.

4. Extract the model anywhere you like on your system.

5. Extra: If you don't have any audio files for testing etc. you can download the samples from DeepSpeech's Releases (the file named audio-0.6.1.tar.gz)

6. When you are done, run this command in your project directory:

   export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$PWD/lib/
   

And done!

// TODO

I will add a bash script for automating this process including the downloading and extracting etc. PRs welcome.

There is a complete example of how to use this module in cmd/main.v

import thecodrr.vspeech
// specify values for use later
const (
    beam_width            = 300
    lm_weight             = 0.75
    valid_word_count_weight = 1.85
)
// create a new model
mut model := vspeech.new("/path/to/the/model.pbmm", 1)

lm := "/path/to/the/lm/file" //its in the models archive
trie := "/path/to/the/trie/file" //its in the models archive
// enable the decoder with language model (optional)
model.enable_decoder_with_lm(lm, trie, lm_weight, valid_word_count_weight)

data := byteptr(0)//raw audio samples (use thecodrr.vave module for this)
data_len := 0 //the total length of the buffer
// convert the audio to text
text := model.speech_to_text(data, data_len)
println(text)

// make sure to free everything
unsafe {
    model.free()
    model.free_string(text)
}

Creates a new Model with the specified model_path and beam_size.

beam_size decides the balance between accuracy and cost. The larger the beam_size the more accurate the decoding will be but at the cost of time and resources.

model_path is the path to the model file. It is the file with .pb extension but it is better to use .pbmm file as it is mmapped and is lighter on the RAM.

The main struct represents the interface to the underlying model. It has the following methods:

Load the Language Model and enable the decoder to use it. Read the method comments to know what each param does.

Use this to get the sample rate expected by the model. The audio samples you need converted MUST match this sample rate.

This is the method that you are looking for. It's where all the magic happens (and also all the bugs).

buffer is the audio data that needs to be decoded. Currently DeepSpeech supports 16-bit RAW PCM audio stream at the appropriate sample rate. You can use thecodrr.vave to read audio samples from a WAV file.

buffer_size is the total number of bytes in the buffer

Same as speech_to_text except this returns a Metadata struct that you can use for output analysis etc.

Create a stream for streaming audio data (from a microphone for example) into the decoder. This, however, isn't an actual stream i.e. there's no seek etc. This will initialize the streaming_statein yourModel` instance which you can use as mentioned below.

Free the Model

Free the string the decoder outputted in speech_to_text.

The streaming state is used to handle pseudo-streaming of audio content into the decoder. It exposes the following methods:

Use this for feeding multiple chunks of data into the stream continuously.

You can use this to get the output of the current data in the stream. However, this is quite expensive due to no streaming capabilities in the decoder. Use this only when necessary.

Call this when streaming is finished and you want the final output of the whole stream.

Same as finish_stream but returns a Metadata struct which you can use to analyze the output.

Call this when done to free the captured StreamingState.

Fields:

items An array of MetadataItems

num_items Total number of items in the items array.

confidence Approximated confidence value for this transcription

Methods:

get_items() - Converts the C pointer MetadataItem array into V array which you can iterate over normally.

get_text() - Helper method to get the combined text from all the MetadataItems outputting the result in one string.

free() - Free the Metadata instance

Fields:

character - The character generated for transcription

timestep - Position of the character in units of 20ms

start_time - Position of the character in seconds

Methods:

str() - Combine and output all the data in the MetadataItem nicely into a string.

Support it by joining stargazers for this repository. ⭐or buy me a cup of coffee And follow me for my next creations! 🤩

MIT License

Copyright (c) 2019 Abdullah Atta

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.